Contribute
Register

Problem with poor Vega 64 Performance on Mojave 10.14.5

Status
Not open for further replies.
@movolans Would you mind sharing your EFI and BIOS settings for the Z390 Designare + Vega64 Nitro+, as you claim to have normal BruceX performance (11-12sec) with 10.14.5?

I'm running an identical setup, and some weeks before I had a 10.14.5 working quite well at around 12 secs, but in the meantime messed it up. Trying to start clean now, but I can't get any better than 16 seconds.

SMBIOS iMac19,1 seems to work better than iMacPro1,1.

I also followed the advice of removing NVME drives as I was getting 20sec+ performance. Now running off a SATA SSD and no other devices than just Vega64 and normal onboard stuff enabled (incl. TB3), the performance is slightly better.

As a quick question as well: How do you exactly count BruceX FCPX performance in time? I Do we start the timer when we hit "Save" or do we start the timer when we hear the GPU start (coilwhile and whatnot)? Do we stop the timer when the coloured pictogram of the file shows up on the desktop, or when QuickTime actually shows the file? There's a few seconds of difference in all these options.

I mean to say, if I'm really optimistic I could claim ca. 12 seconds, but it doesn't feel like I'm counting the whole process then.

As far as I understand now, with either SMBIOS one doesn't need to provide Device Properties or ACPI patches, as Whatevergreen should take care of this?


PS: If people report that they have good performance in 10.14.3, but it changes in 10.14.5, you can exclude any hardware intereference, like sharing of PCI-E bandwidth. This is all defined on a low level in the BIOS, rather than by the OS.

Hiya.

Unfortunately my system has changed pretty significantly since my original post. I have switched to a Z390 Designare MoBo and Radeon VII GPU. I still get around 13 seconds. Things really changed for my me on my old board when I removed the NVME just above the PCIe 16X slot. On this MoBo I don't have the same issue - both NVME slots are populated.

I'm convinced this was only part of the issue (and may have been very specific to my system) drivers not being optimised and similar issues in FCPX are probably also to blame. Hopefully this will change in future updates.

I've also switched to Davinci Resolve Studio and I can tell you that exports / renders are lightning fast. I just did an H264 export of a 3 minute video in about 20 seconds :) So the performance is there - just not across the board.
 
9900k with Vega56 about 13s
 
Hiya.

Unfortunately my system has changed pretty significantly since my original post. I have switched to a Z390 Designare MoBo and Radeon VII GPU. I still get around 13 seconds. Things really changed for my me on my old board when I removed the NVME just above the PCIe 16X slot. On this MoBo I don't have the same issue - both NVME slots are populated.

I'm convinced this was only part of the issue (and may have been very specific to my system) drivers not being optimised and similar issues in FCPX are probably also to blame. Hopefully this will change in future updates.

I've also switched to Davinci Resolve Studio and I can tell you that exports / renders are lightning fast. I just did an H264 export of a 3 minute video in about 20 seconds :) So the performance is there - just not across the board.

Hey, as a matter of fact I have the Z390 Designare now as well, so we're equal on that. I had two NVME drives and a PCI-E1x WiFi chip attached, and in 10.14.5 this resulted in 20 sec BruceX export time. After removing all extra's, the result became better, 14~16 sec. After installing Catalina beta 2, it's all normal with 12 seconds now. I'm not sure if the problem is specifically in 10.14.5, but it seems that way.
 
Things really changed for my me on my old board when I removed the NVME just above the PCIe 16X slot. On this MoBo I don't have the same issue - both NVME slots are populated.

I'm convinced this was only part of the issue (and may have been very specific to my system) drivers not being optimised and similar issues in FCPX are probably also to blame. Hopefully this will change in future updates.


@movolans,

All Intel Desktop class CPU's only have 16 PCIe lanes to the PCI Bus, its possible on your older motherboard that your NVMe was in a M.2 slot that was directly connected to the CPU PCIe bus rather than going through the PCH as such its possible that your dGPU was only using x8 PCIe lanes rather than the full x16 PCIe lanes as the M.2 NVMe was using some of them. Once you removed the NVMe then the dGPU was able to to use the full x16 bandwidth resulting in improved dGPU performance.

In most applications the dGPU running in a x8 configuration rather than the full x16 does not impact significantly on performance however it does depend on the application and i suspect that video encode is probably most effected.

Some motherboards use PCIe lane switches to allow more than one device to share the x16 lanes to the CPU, other motherboards route the M.2 slots through the PCH which means that they will share the x4 PCH to CPU PCie lanes with all the other the devices connected to the PCH.

Most motherboards have a block diagram in the manual which shows which devices/PCIe slots/M.2 Slots are directly connected to the CPU and which are routed through the PCH.

Its always a good idea to analyse these diagrams so that you put your devices in the most appropriate slots.

Cheers
Jay
 
Last edited:
@movolans,

All Intel Desktop class CPU's only have 16 PCIe lanes to the PCI Bus, its possible on your older motherboard that your NVMe was in a M.2 slot that was directly connected to the CPU PCIe bus rather than going through the PCH as such its possible that your dGPU was only using x8 PCIe lanes rather than the full x16 PCIe lanes as the M.2 NVMe was using some of them. Once you removed the NVMe then the dGPU was able to to use the full x16 bandwidth resulting in improved dGPU performance.

Some motherboards use PCIe lane switches to allow more than one device to share the x16 lanes to the CPU, other motherboards route the M.2 slots through the PCH which means that they will share the x4 PCH to CPU PCie lanes with all the other the devices connected to the PCH.

Most motherboards have a block diagram in the manual which shows which devices/PCIe slots/M.2 Slots are directly connected to the CPU and which are routed through the PCH.

Its always a good idea to analyse these diagrams so that you put your devices in the most appropriate slots.

Cheers
Jay

Thanks Jay aware of all of this.

Seems that the Designare topology is preferable to the Gaming 7 in this case.

I suspect we’ll see performance increase as Apple improve optimisation. A little early I the day for Catalina beta for me though. I’ll hang on.
 
I suspect we’ll see performance increase as Apple improve optimisation. A little early I the day for Catalina beta for me though. I’ll hang on.


@movolans,

Apple tend to add functionality first in early beta's and then concentrate on optimisations in later bata's and minor os releases once a major os update is released.

Right now i'm on the fence with Catalina, i have quite a few 32 bit apps that i use daily so for now i'm sticking with Mojave, hopefully those apps will be updated to 64 bit versions by the time Catalina is released proper.

I think Mojave will be a milestone OS for many people as its still supports 32Bit Apps as well as Open GL/CL but is capable of still running Metal 2 Apps, from what i've seen of Catalina iso far t offers no real new features that i'm interested in

Cheers
Jay
 
Last edited:
@movolans,

Apple tend to add functionality first in early beta's and then concentrate on optimisations in later bata's and minor os releases once a major os update is released.

Right now i'm on the fence with Catalina, i have quite a few 32 bit apps that i use daily so for now i'm sticking with Mojave, hopefully those apps will be updated to 64 bit versions by the time Catalina is released proper.

I think Mojave will be a milestone OS for many people as its still supports 32Bit Apps as well as Open GL/CL but is capable of still running Metal 2 Apps, from what i've seen of Catalina iso far t offers no real new features that i'm interested in

Cheers
Jay

Yep that’s the way they’ve done it forever

I’m certainly not rushing into Catalina. I’m usually a super early adopter but think I’ll take it easy on this one.
 
My Sapphire RX Vega 64 (LE) running on 10.14.5 LuxMark ended with 30402 score
 
Status
Not open for further replies.
Back
Top